Relational Scaffolds in Novel Learning and Decision Making
How do intelligent agents make decisions in novel environments? The generalization of knowledge to novel circumstances relies on the ability to form structural representations invariant to specific features of the input domain. This means basis sets thought previously to be evolutionarily hard-coded to represent magnitude or cardinality and the Euclidean dimensions of the external world (e.g. space, time, number) may instead reflect a natural set of low-dimensional basis functions that supports
